Publish curated Unraid-AppOps-Gateway source from ae6b891c48da412de613ce0ed52c0bc1e9908fbf

PUBLIC_SOURCE_EXPORT.md

@@ -1,3 +1,3 @@
 # Public source export
 
-This source snapshot was generated from the private canonical repository at revision `8e49e71ad0298326f565f57d4962e482e738d6e1`. It contains no private Git history or operational evidence. Changes are published from the canonical repository.
+This source snapshot was generated from the private canonical repository at revision `ae6b891c48da412de613ce0ed52c0bc1e9908fbf`. It contains no private Git history or operational evidence. Changes are published from the canonical repository.

PUBLIC_SOURCE_MANIFEST.json

@@ -1,6 +1,6 @@
 {
   "schemaVersion": 1,
-  "sourceRevision": "8e49e71ad0298326f565f57d4962e482e738d6e1",
+  "sourceRevision": "ae6b891c48da412de613ce0ed52c0bc1e9908fbf",
   "files": [
     {
       "path": ".dockerignore",
@@ -1659,8 +1659,8 @@
     },
     {
       "path": "backend/tests/test_runner_fleet_specialize.py",
-      "sha256": "e3134a8ad84ab23fa2fef3d377d25b891cc4d854b8bdad7f271048404fbfc249",
+      "sha256": "5bd85d16db6ef824a8c83bdafb4e1a70cbb736064611929e538df7224748f885",
-      "bytes": 205429
+      "bytes": 205798
     },
     {
       "path": "backend/tests/test_runner_fleet_v2_execution_surface.py",
@@ -1684,8 +1684,8 @@
     },
     {
       "path": "backend/tests/test_runner_resource_envelope.py",
-      "sha256": "f2a1c2468bdda62acbc91657a35cfc63fe525413ee56dc66b8bd4527597c9aba",
+      "sha256": "dfba7e7ba84689205ed677a22aa2f7daca1898def800a9e7b6fa60528a5d0b73",
-      "bytes": 101609
+      "bytes": 116563
     },
     {
       "path": "backend/tests/test_runner_scheduler_liveness_recovery.py",

backend/tests/test_runner_fleet_specialize.py

@@ -23,10 +23,11 @@ import pytest
     "resource_cgroup_attestation_failed", "resource_cgroup_v2_unavailable",
     "Bearer secret-value", None,
 ])
-def test_build_resource_refusal_keeps_only_allowlisted_private_cause(specialize, monkeypatch, reason):
+@pytest.mark.parametrize("sample_stage", ["after_snapshot", "Bearer secret-value", {"secret": "value"}, None])
+def test_build_resource_refusal_keeps_only_allowlisted_private_cause(specialize, monkeypatch, reason, sample_stage):
     monkeypatch.setattr(specialize, "_docker_outer_inventory", lambda: ([], []))
     monkeypatch.setattr(specialize, "attest_build_probe_workload", lambda *_a, **_k: {
-        "ready": False, "reason": reason,
+        "ready": False, "reason": reason, "sampleStage": sample_stage,
     })
     with pytest.raises(specialize.RunnerFleetSpecializeError) as error:
         specialize._build_probe_resource_attestation({"containerId": "a" * 64}, {"State": {"Pid": 77}})
@@ -45,6 +46,10 @@ def test_build_resource_refusal_keeps_only_allowlisted_private_cause(specialize,
     }, error.value)
     assert len(records) == 1
     assert records[0]["causeCode"] == (reason if reason in specialize._RESOURCE_FAILURE_CODES else str(error.value))
+    if reason == "resource_cgroup_process_mapping_invalid" and sample_stage == "after_snapshot":
+        assert records[0]["sampleStage"] == "after_snapshot"
+    else:
+        assert "sampleStage" not in records[0]
     assert "secret-value" not in json.dumps(records)
 
 

backend/tests/test_runner_resource_envelope.py

@@ -1286,6 +1286,216 @@ def _owned_family_fixture(monkeypatch, *, exact=False, init=True, docker=True):
     return envelope, claim, rows, processes, controls, writes, inodes
 
 
+@pytest.mark.parametrize("case", ["owned", "foreign_zero", "duplicate", "escaped_thread", "missing_thread", "family_process"])
+def test_coherent_namespace_sample_requires_all_threads_contained(monkeypatch, case):
+    envelope, claim, _rows, processes, _controls, writes, _inodes = _owned_family_fixture(monkeypatch, exact=True)
+    leaf = f"/itworx-ci/{claim['containerId']}"
+    if case == "foreign_zero":
+        processes[leaf + "/init"] = "1 0"
+    elif case == "duplicate":
+        processes[leaf + "/init"] = "1 1"
+    elif case == "family_process":
+        processes["/itworx-ci"] = "77"
+    tids = (1, 2, 9) if case == "escaped_thread" else (1,) if case == "missing_thread" else (1, 2)
+    monkeypatch.setattr(envelope, "_probe_proc_task_ids", lambda _pids: (tids, ((1, 36, 901), (2, 36, 902))))
+    if case == "owned":
+        sample = envelope._probe_namespace_thread_membership(101, leaf)
+        assert sample["processIds"] == (1, 2)
+        assert sample["threadIds"] == (1, 2)
+    else:
+        reason = "resource_cgroup_probe_namespace_invalid" if case == "family_process" else "resource_cgroup_process_mapping_invalid"
+        with pytest.raises(RuntimeError, match="^" + reason + "$"):
+            envelope._probe_namespace_thread_membership(101, leaf)
+    assert writes == []
+
+
+def test_build_snapshot_helper_is_fixed_dind_namespace_and_readonly_family():
+    envelope = _module()
+    target = {"name": envelope.DIND_NAME, "id": "a" * 64}
+    args = envelope._host_probe_create_args("build_snapshot", "sha256:" + "b" * 64,
+                                           envelope._HOST_VOLUME_ROOT + "/controller-artifacts/" + "c" * 40,
+                                           namespace_target=target)
+    assert "container:" + target["id"] in args
+    assert "type=bind,source=/sys/fs/cgroup/itworx-ci,target=/host-cgroup,readonly,bind-propagation=rprivate" in args
+    assert not any("/proc" in value or "--pid=host" in value or "docker.sock" in value for value in args)
+    with pytest.raises(RuntimeError, match="resource_cgroup_probe_namespace_invalid"):
+        envelope._host_probe_create_args("build_snapshot", "sha256:" + "b" * 64,
+                                         envelope._HOST_VOLUME_ROOT + "/controller-artifacts/" + "c" * 40,
+                                         namespace_target={"name": "itworx-ci-checks", "id": "a" * 64})
+
+
+@pytest.mark.parametrize("case", ["owned", "thread_race_once", "process_count", "inode_drift", "limit_drift", "wrong_pid", "running_outer", "boolean_pid", "fractional_pid", "fractional_cpu", "fractional_memory", "malformed_name"])
+def test_coherent_build_snapshot_retries_only_inconsistent_membership(monkeypatch, case):
+    envelope, claim, rows, processes, _controls, writes, inodes = _owned_family_fixture(monkeypatch, exact=True)
+    leaf = f"/itworx-ci/{claim['containerId']}"
+    probe_id = "b" * 64
+    path = leaf + "/docker/" + probe_id
+    rows[path] = {"path": path, "cpuMax": "400000 100000", "memoryMax": str(6 * 1024**3)}
+    processes[leaf + "/init"] = "1"
+    old_open, old_read, old_stat = envelope._probe_open_relative, envelope._probe_read, envelope.os.fstat
+    monkeypatch.setattr(envelope, "_probe_open_relative", lambda fd, name: 909 if name == path else old_open(fd, name))
+    monkeypatch.setattr(envelope, "_probe_read", lambda fd, name: "2" if fd == 909 and name in {"cgroup.procs", "cgroup.threads"} else old_read(fd, name))
+    monkeypatch.setattr(envelope.os, "fstat", lambda fd: SimpleNamespace(st_dev=36, st_ino=909) if fd == 909 else old_stat(fd))
+    proc_calls, snapshots = [], []
+
+    def proc(_pids):
+        assert _pids == (1, 2)
+        proc_calls.append(True)
+        tids = (1, 2, 9) if case == "thread_race_once" and len(proc_calls) == 1 else (1, 2)
+        return tids, ((1, 36, 901), (2, 36, 902))
+
+    def snapshot(payload, _fd):
+        snapshots.append(payload)
+        result = {"version": 2, "groups": [dict(row) for row in rows.values()]}
+        if case == "inode_drift":
+            inodes[leaf + "/init"] += 1
+        elif case == "limit_drift":
+            rows[leaf]["cpuMax"] = "500000 100000"
+        return result
+
+    monkeypatch.setattr(envelope, "_probe_proc_task_ids", proc)
+    monkeypatch.setattr(envelope, "_probe_snapshot", snapshot)
+    monkeypatch.setattr(envelope.time, "sleep", lambda _seconds: None)
+    outer = [{"name": profile.container_name, "id": format(index, "064x"), "running": False}
+             for index, profile in enumerate(envelope.RUNNER_PROFILES.values(), 1)]
+    outer.append({"name": envelope.DIND_NAME, "id": claim["containerId"], "running": True})
+    if case == "running_outer":
+        outer[0]["running"] = True
+    request = {"op": "build_snapshot", "target": {"name": envelope.DIND_NAME, "id": claim["containerId"]},
+               "outer": outer, "processCount": 3 if case == "process_count" else 2,
+               "workloads": [{"id": probe_id, "nestedPid": 77 if case == "wrong_pid" else 2,
+                              "cpuMilli": 4000, "memoryBytes": 6 * 1024**3}]}
+    for mutation, key, value in (("boolean_pid", "nestedPid", True), ("fractional_pid", "nestedPid", 2.0),
+                                 ("fractional_cpu", "cpuMilli", 4000.0), ("fractional_memory", "memoryBytes", float(6 * 1024**3))):
+        if case == mutation:
+            request["workloads"][0][key] = value
+    if case == "malformed_name":
+        outer[0]["name"] = []
+    if case in {"owned", "thread_race_once"}:
+        result = envelope._probe_coherent_build_snapshot(request, 101)
+        assert result["namespaceProof"] == {"allThreadsContained": True, "processCount": 2, "threadCount": 2}
+        assert len(snapshots) == 1
+        assert len(proc_calls) == (3 if case == "thread_race_once" else 2)
+    else:
+        expected = {"process_count": "process_mapping_invalid", "inode_drift": "probe_namespace_invalid",
+                    "limit_drift": "probe_namespace_invalid", "wrong_pid": "probe_workload_mismatch",
+                    "running_outer": "probe_request_invalid", "boolean_pid": "probe_request_invalid",
+                    "fractional_pid": "probe_request_invalid", "fractional_cpu": "probe_request_invalid",
+                    "fractional_memory": "probe_request_invalid", "malformed_name": "probe_request_invalid"}[case]
+        with pytest.raises(RuntimeError, match="^resource_cgroup_" + expected + "$"):
+            envelope._probe_coherent_build_snapshot(request, 101)
+        assert len(snapshots) == (1 if case in {"inode_drift", "limit_drift"} else 0)
+        if case == "process_count":
+            assert len(proc_calls) == 3
+    assert writes == []
+
+
+@pytest.mark.parametrize("drift", ["none", "threads", "process", "identity"])
+def test_build_coherent_sample_keeps_complete_binding_and_exact_process_ids(monkeypatch, drift):
+    envelope = _module()
+    original = ({"Id": "a" * 64, "StartedAt": "start", "allIdentityFields": {"memory": 12 * 1024**3}}, [(10, 2)])
+    current = copy.deepcopy(original)
+    if drift == "threads":
+        current[1][0] = (10, 3)
+    elif drift == "process":
+        current[1][0] = (11, 2)
+    elif drift == "identity":
+        current[0]["allIdentityFields"]["memory"] = 0
+    monkeypatch.setattr(envelope, "_engine_outer_population", lambda _name: current)
+    if drift in {"none", "threads"}:
+        envelope._assert_outer_process_ids_unchanged(envelope.DIND_NAME, original)
+    else:
+        reason = "resource_cgroup_probe_namespace_invalid" if drift == "identity" else "resource_cgroup_process_mapping_invalid"
+        with pytest.raises(RuntimeError, match="^" + reason + "$"):
+            envelope._assert_outer_process_ids_unchanged(envelope.DIND_NAME, original)
+
+
+@pytest.mark.parametrize("drift", ["none", "host_pid", "foreign_cid", "capability", "writable", "host_proc", "source"])
+def test_build_snapshot_inspection_retains_exact_private_boundary(drift):
+    envelope = _module()
+    item, image, source = _host_probe_fixture(envelope)
+    target = {"name": envelope.DIND_NAME, "id": "a" * 64}
+    item["HostConfig"]["PidMode"] = "container:" + target["id"]
+    item["Config"]["Labels"].update({"tech.itworx.probe.namespace-name": target["name"], "tech.itworx.probe.namespace-id": target["id"]})
+    if drift == "host_pid":
+        item["HostConfig"]["PidMode"] = "host"
+    elif drift == "foreign_cid":
+        item["HostConfig"]["PidMode"] = "container:" + "f" * 64
+    elif drift == "capability":
+        item["HostConfig"]["CapAdd"] = ["DAC_OVERRIDE"]
+    elif drift == "writable":
+        item["Mounts"][0]["RW"] = True
+    elif drift == "host_proc":
+        item["Mounts"].append({"Type": "bind", "Source": "/proc", "Destination": "/host-proc", "RW": False})
+    elif drift == "source":
+        item["Mounts"][1]["Source"] += "/foreign"
+    assert envelope._validate_host_probe_inspect(item, mode="build_snapshot", image_id=image, code_source=source,
+                                                namespace_target=target) is (drift == "none")
+
+
+@pytest.mark.parametrize("case", ["owned", "missing_proof", "false_proof", "foreign_count", "boolean_count", "extra_field", "running_outer", "wrong_workload", "identity_drift", "pid_drift", "late_runner_start", "late_runner_replaced"])
+def test_collect_build_snapshot_requires_complete_coherent_proof(monkeypatch, case):
+    envelope = _module()
+    snapshot, nested_items, _calls = _install_classic_builder_collection(envelope, monkeypatch, {})
+    nested_items[:] = nested_items[:1]
+    nested_items[0]["HostConfig"].update(NanoCpus=4_000_000_000, Memory=6 * 1024**3)
+    original_inspect = envelope._inspect_container
+    helper_completed = []
+
+    def inspect(name):
+        item = copy.deepcopy(original_inspect(name))
+        if name != envelope.DIND_NAME:
+            item["State"].update(Running=case == "running_outer" and name == "itworx-ci-build", Pid=0)
+            if helper_completed and name == "itworx-ci-build":
+                if case == "late_runner_start":
+                    item["State"].update(Running=True, Pid=123)
+                elif case == "late_runner_replaced":
+                    item["Id"] = "f" * 64
+        return item
+
+    monkeypatch.setattr(envelope, "_inspect_container", inspect)
+    binding = envelope._engine_outer_binding(inspect(envelope.DIND_NAME))
+    populations, probes = [], []
+
+    def population(_name):
+        populations.append(True)
+        current = copy.deepcopy(binding)
+        if len(populations) > 1 and case == "identity_drift":
+            current["Id"] = "f" * 64
+        return current, [(11 if len(populations) > 1 and case == "pid_drift" else 10, len(populations))]
+
+    def probe(operation, payload):
+        helper_completed.append(True)
+        probes.append((operation, payload))
+        assert operation == "build_snapshot"
+        assert payload["target"] == {"name": envelope.DIND_NAME, "id": binding["Id"]}
+        assert payload["processCount"] == 1
+        result = {"snapshot": snapshot, "namespaceProof": {"allThreadsContained": True, "processCount": 1, "threadCount": 2}}
+        if case == "missing_proof":
+            del result["namespaceProof"]
+        elif case == "false_proof":
+            result["namespaceProof"]["allThreadsContained"] = False
+        elif case in {"foreign_count", "boolean_count"}:
+            result["namespaceProof"]["processCount"] = 2 if case == "foreign_count" else True
+        elif case == "extra_field":
+            result["namespaceProof"]["extra"] = True
+        return result
+
+    monkeypatch.setattr(envelope, "_engine_outer_population", population)
+    monkeypatch.setattr(envelope, "_run_host_probe", probe)
+    candidate = ("f" * 64 if case == "wrong_workload" else nested_items[0]["Id"], nested_items[0]["State"]["Pid"])
+    if case == "owned":
+        assert envelope._collect_snapshot(envelope.FIXED_RUNNER_AND_DIND_NAMES, coherent_build_probe=candidate) == snapshot
+        assert len(probes) == 1 and len(populations) == 2
+    else:
+        reason = "resource_cgroup_probe_workload_mismatch" if case in {"running_outer", "wrong_workload"} else (
+            "resource_cgroup_probe_namespace_invalid" if case in {"identity_drift", "late_runner_start", "late_runner_replaced"} else
+            "resource_cgroup_process_mapping_invalid" if case == "pid_drift" else "resource_cgroup_probe_snapshot_invalid")
+        with pytest.raises(RuntimeError, match="^" + reason + "$"):
+            envelope._collect_snapshot(envelope.FIXED_RUNNER_AND_DIND_NAMES, coherent_build_probe=candidate)
+        assert len(probes) == (0 if case in {"running_outer", "wrong_workload"} else 1)
+
+
 @pytest.mark.parametrize("init,docker,exact", [(False, False, False), (True, False, False), (True, True, False), (True, True, True)])
 def test_owned_dind_family_binds_only_existing_exact_private_child(monkeypatch, init, docker, exact):
     envelope, claim, _rows, _processes, _controls, writes, _inodes = _owned_family_fixture(
@@ -1965,7 +2175,7 @@ def test_build_probe_attestor_binds_exact_cid_nested_pid_and_build_cgroup(monkey
     attempts = []
     waits = []
 
-    def collect(_expected):
+    def collect(_expected, **_kwargs):
         attempts.append(True)
         if len(attempts) <= race_count:
             raise RuntimeError("resource_cgroup_process_mapping_invalid")
@@ -2006,7 +2216,7 @@ def test_build_probe_attestor_does_not_retry_other_resource_refusals(monkeypatch
     envelope = _module()
     attempts = []
 
-    def collect(_expected):
+    def collect(_expected, **_kwargs):
         attempts.append(True)
         raise RuntimeError(reason)
 
@@ -2017,7 +2227,36 @@ def test_build_probe_attestor_does_not_retry_other_resource_refusals(monkeypatch
     assert len(attempts) == 1 and "probeWorkload" not in proof
 
 
-@pytest.mark.parametrize("drift", [None, "binding_before_snapshot", "binding_after_snapshot", "threads_after_snapshot", "binding_and_threads_after_snapshot"])
+@pytest.mark.parametrize("stage", [
+    "before_helper_start", "after_helper_exit", "namespace_counts",
+    "host_population_counts", "host_snapshot_counts", "outer_population_compare",
+    "after_snapshot", "Bearer secret-value", {"secret": "value"}, None,
+])
+def test_build_population_refusal_retains_only_finite_sample_stage(monkeypatch, stage):
+    envelope = _module()
+    attempts = []
+    sleeps = []
+
+    def refuse(_names, **_kwargs):
+        attempts.append(True)
+        error = RuntimeError("resource_cgroup_process_mapping_invalid")
+        error.sample_stage = stage
+        raise error
+
+    monkeypatch.setattr(envelope, "_collect_snapshot", refuse)
+    monkeypatch.setattr(envelope.time, "sleep", sleeps.append)
+    result = envelope.attest_build_probe_workload("a" * 64, 77)
+    assert result["ready"] is False
+    assert result["reason"] == "resource_cgroup_process_mapping_invalid"
+    assert len(attempts) == 3 and sleeps == [0.1, 0.1]
+    if isinstance(stage, str) and stage in envelope.POPULATION_SAMPLE_STAGES:
+        assert result["sampleStage"] == stage
+    else:
+        assert "sampleStage" not in result
+    assert "secret-value" not in json.dumps(result)
+
+
+@pytest.mark.parametrize("drift", [None, "binding_before_snapshot", "binding_after_snapshot", "threads_after_snapshot", "binding_and_threads_after_snapshot", "inventory_thread_churn", "inventory_pid_churn", "inventory_identity_drift"])
 def test_collect_snapshot_uses_host_helper_and_exports_real_family_counters(monkeypatch, drift):
     envelope = _module()
     snapshot = _snapshot()
@@ -2064,12 +2303,15 @@ def test_collect_snapshot_uses_host_helper_and_exports_real_family_counters(monk
         }
 
     docker_commands = []
+    inventory_collected = False
 
     def command(args, **_kwargs):
+        nonlocal inventory_collected
         docker_commands.append(args)
         if "info" in args:
             return "2|cgroupfs|true" if "exec" in args else "2|cgroupfs"
         if "docker" in args and "ps" in args:
+            inventory_collected = True
             return ""
         raise AssertionError(f"unexpected fixed Docker command: {args}")
 
@@ -2090,6 +2332,17 @@ def test_collect_snapshot_uses_host_helper_and_exports_real_family_counters(monk
         namespace_calls[name] = namespace_calls.get(name, 0) + 1
         binding, pairs, claim = copy.deepcopy(stable_namespace(name))
         if name == envelope.DIND_NAME:
+            # Docker exec used by inventory can start/retire daemon threads or
+            # processes. Measure only after our own inventory commands finish;
+            # a changed container identity must still refuse before sampling.
+            if inventory_collected and drift == "inventory_thread_churn":
+                pairs[0] = (pairs[0][0], 2)
+                claim["threadCount"] = 2
+            elif inventory_collected and drift == "inventory_pid_churn":
+                pairs.append((pairs[0][0] + 1000, 1))
+                claim.update(processCount=2, threadCount=2)
+            elif inventory_collected and drift == "inventory_identity_drift":
+                binding["State"]["StartedAt"] = "changed-start"
             phase = "before_snapshot" if namespace_calls[name] == 1 else "after_snapshot"
             if drift and drift.endswith(phase):
                 if drift.startswith("binding"):
@@ -2113,7 +2366,12 @@ def test_collect_snapshot_uses_host_helper_and_exports_real_family_counters(monk
         expected_host_budget={"cpuMilli": 8000, "memoryMiB": 16384},
     )
 
-    if drift:
+    if drift == "inventory_identity_drift":
+        assert result["ready"] is False
+        assert result["reason"] == "resource_cgroup_probe_namespace_invalid"
+        assert probe_calls == []
+        return
+    if drift and not drift.startswith("inventory_"):
         expected = "resource_cgroup_probe_namespace_invalid" if drift.startswith("binding") else "resource_cgroup_process_mapping_invalid"
         assert result["ready"] is False and result["reason"] == expected
         if drift.endswith("before_snapshot"):